Recent advances and application of machine learning in food flavor prediction and regulation

Background: Food flavor is a key factor affecting sensory quality. Predicting and regulating flavor can result in exceptional flavor characteristics and improve consumer preferences and food acceptability. Evaluating and regulating flavor through traditional experimental methods are time-consuming, labor-intensive, and cannot handle large amounts of data. Computational methods, such as machine learning (ML) techniques, can accurately and efficiently predict and regulate complex flavors and attract continuous attention. Scope and approach: This review presents the principles and advantages of commonly used ML methods, including support vector machine, decision tree, random forest, k-nearest neighbors, extreme learning machine, artificial neural networks, and deep learning, as well as their recent applications and prospects in the prediction and regulation of food flavors. Notably, the prediction of food flavor based on molecular structures, physical and chemical properties, and data obtained from electronic nose, electronic tongue, and gas chromatography-mass spectrometry were summarized. The regulation of food flavor by ML through metabolites and genes has also been reviewed. Key findings and conclusions: Simultaneous combination of various ML methods could improve the prediction accuracy of flavor profiles, perception intensity, and sensory quality classification compared to a single model. Additionally, the data fusion of different techniques showed better flavor prediction performance than single data input. This review indicates that ML techniques are promising for predicting flavor formation mechanisms, dose effects of structure-flavor quality, and directing the bio/chemical synthesis of desirable flavor compounds to meet the consumer demand for healthy and delicious food.